CN109766525A - A kind of sensitive information leakage detection framework of data-driven - Google Patents
A kind of sensitive information leakage detection framework of data-driven Download PDFInfo
- Publication number
- CN109766525A CN109766525A CN201910030202.3A CN201910030202A CN109766525A CN 109766525 A CN109766525 A CN 109766525A CN 201910030202 A CN201910030202 A CN 201910030202A CN 109766525 A CN109766525 A CN 109766525A
- Authority
- CN
- China
- Prior art keywords
- text
- fingerprint
- data
- sensitive information
- information
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 36
- 238000004458 analytical method Methods 0.000 claims abstract description 39
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 230000009467 reduction Effects 0.000 claims abstract description 10
- 239000000284 extract Substances 0.000 claims description 15
- 230000035945 sensitivity Effects 0.000 claims description 8
- 238000012544 monitoring process Methods 0.000 claims description 4
- 230000000877 morphologic effect Effects 0.000 claims description 4
- 230000006798 recombination Effects 0.000 claims description 4
- 238000005215 recombination Methods 0.000 claims description 4
- 230000003287 optical effect Effects 0.000 claims description 2
- 238000000034 method Methods 0.000 abstract description 22
- 238000012545 processing Methods 0.000 description 14
- 230000008569 process Effects 0.000 description 11
- 230000005540 biological transmission Effects 0.000 description 7
- 238000005516 engineering process Methods 0.000 description 7
- 230000011218 segmentation Effects 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 5
- 238000007689 inspection Methods 0.000 description 5
- 238000012015 optical character recognition Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 3
- 230000004044 response Effects 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 238000010276 construction Methods 0.000 description 2
- 239000012634 fragment Substances 0.000 description 2
- 238000011430 maximum method Methods 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 230000006399 behavior Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 230000006378 damage Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005538 encapsulation Methods 0.000 description 1
- 230000012010 growth Effects 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 230000008140 language development Effects 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 238000011946 reduction process Methods 0.000 description 1
- 230000010076 replication Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
Landscapes
- Storage Device Security (AREA)
Abstract
The present invention relates to sensitive information leakage detection fields.Specifically related to the method for finding sensitive information leakage by capture, analysis and reduction to network packet etc..The invention proposes a kind of sensitive information leakage detection framework of data-driven, which is mainly made of five parts: information collection module, Text Feature Extraction module, fingerprint generation module, fingerprint database, text analysis model.Information collection module first captures the data packet on the portal, the original document in data packet is restored later, then Text Feature Extraction module carries out the extraction of content of text to these files, the finger print information of fingerprint generation module generation corresponding document, the fingerprint of sensitive information in the finger print information and fingerprint database of generation can be carried out text similarity calculating by last text analysis model, to judge whether sensitive information leakage occurred.Its structure is as shown in Figure 1.The frame can be quickly detected from sensitive information leakage and promptly respond.
Description
Technical field
The present invention relates to sensitive information leakage detection fields.Specifically related to the capture passed through to network packet, analyze
And reduction etc. is come the method that finds sensitive information leakage.
Background technique
Sensitive information suffers from each moment in stage of the life cycles such as its creation, storage, transmission, use, destruction
The risk of leakage has that number of values must be furtherd investigate.With the rapid development of internet, sensitive information faces more
Cyberthreat.In network and information facility layer, private network, public network, calculating and storage facility are faced with a large amount of attack prestige
The side of body, once network and infrastructure are kidnapped by attack and would potentially result in sensitive information leakage.Furthermore high speed network bandwidth is sharply
Growth also brings huge difficulty to the detection of sensitive information leakage.
Technology currently used for detecting sensitive information leakage mainly uses static analysis and dynamic analysis both modes, quiet
State analysis is mostly used in loophole present in offline inspection application program, cannot execution state to application program carry out it is real-time
Monitoring, thus the sensitive information leakage that application program can not occur detects.Meanwhile the result foot in order to make static analysis
It is enough accurate, it needs to carry out a large amount of priori test in advance.Dynamic analysis mainly use dynamic stain tracking technique to sensitive information
Leakage is detected.This mode can carry out offline inspection with assisted static analysis, at the same can real-time detection, but which is deposited
The problems such as operational efficiency is low, complexity is high.The method prevailing scenario of traditional detection sensitive information leakage is to be directed to answer
With program, potential loophole present in application program is found by analysis method either statically or dynamically, thus patching bugs
Sensitive information leakage is avoided, is capable of the seldom of real-time tracking sensitive data.In addition under high speed network environment of today, exist
A large amount of network data, traditional detection method can not timely detect data, generally be taken to corresponding data
Switch to offline inspection after being stored, this can seriously affect the timeliness of detection.
In filed of network information security, Intrusion Detection Technique can be effectively prevented as one of key technology therein
Leakage problem occurs for sensitive information.Intrusion Detection Technique is broadly divided into Host Intrusion Detection System and network-based invasion inspection
Survey two major classes.Most of intrusion detection product belongs to Network Intrusion Detection System, the intrusion detection system of the type at present
The data source of system is the real time data packet in network.The system can place on arbitrary node in a network, and selection is different
Placement location be suitable for different network structures, can monitor network flow in real time, check network state and other suspicious work
It is dynamic.In such intruding detection system, it is usually promiscuous mode by the network interface card mode setting of certain host, monitors and examine
Survey all network packets flowed through in this network segment.The intruding detection system of the type can be protected all in entire network segment
Host.It monitors the various data packets in this network segment in real time, to by all data packets of this network segment carry out feature extraction with
Analysis, compares whether packet content matches with the rule in rule base.If the two matches, which just sentences
Surely it is invaded, while being sounded an alarm according to default setting and taking responsive measures.But Intrusion Detection Technique depends critically upon
The update of rule base, since the rule in rule base can not cope with all situations, for certain novel Network Intrusion meeting
There is the problem of intrusion detection failure, while the problem of the technology can not be effectively prevented sensitive information internal leakage.
Summary of the invention
Real-time and the low problem of operational efficiency existing for method of the present invention for traditional detection sensitive information leakage,
And the shortcomings that existing method, propose a kind of sensitive information leakage detection framework of data-driven.The frame can be to network
Data packet carries out monitoring in real time and analysis, and can make an immediate response to the leakage of sensitive information, improves the efficiency of detection;
There is good support to the sensitive information leakage detection under high speed network environment simultaneously, fully solves to hold in this process
The problems such as packet loss that easily occurs, data packet processing speed are slow and cannot judge response in time.Particular content is as follows:
1. frame brief introduction
The specific implementation structure of the frame five parts as shown in Figure 1, be mainly made of: information collection module, Text Feature Extraction
Module, fingerprint generation module, fingerprint database, text analysis model.Information collection module is first to the number on the portal
It is captured according to packet, and layer-by-layer protocol analysis and analysis is carried out to the data packet being collected into, restore the original text in data packet
Part, then Text Feature Extraction module carries out the extraction of content of text to these original documents, and fingerprint generation module is corresponding for generating
The finger print information of file, last text analysis model can be the sensitive informations deposited in the finger print information and fingerprint database of generation
Fingerprint carry out text similarity calculating, to judge whether sensitive information leakage occurred.
Modules have used unified data stream programming model to design, and specifically include input channel, output channel, meter
The components such as calculation task and running environment, as shown in Figure 2.Data flow model is a kind of high performance parallel program by data-driven
Model is executed, executing model can indicate that each node in figure is expressed as a calculation block, for executing with digraph
Calculating task, each edge are expressed as a data transmission channel, are connected between adjacent node by data transmission channel, previous section
Point treated input data of the data as next calculation block, to realize the separation of data processing and data transmission.
Compared with traditional multi-thread programming model, each node in data flow model is all an independent calculating
Block exists independently of other any nodes, will not interfere each other.After this execution model allows input data to reach node
Be immediately performed, due to there is no data dependency between each other in the whole process, the problems such as being not in deadlock, without management and
The concurrent problem such as thread or processing semaphore is manually generated, the complexity of exploitation is reduced.Therefore modules are without concentrating control
System, for any calculation block therein, as long as input has available resources, which can normally execute calculating task, institute
Easily all available concurrencys can be made full use of in each stage.Inside modules, calculating task with
Running environment constitutes a calculation block, and input, output channel composition data transmission channel, each specific calculating task of module is such as
Under:
2. information collection module
Information collection module is mainly responsible for and captures to the data packet transmitted on the portal, to complete to letter
The collection work in breath source.It can specify special network interface or be promiscuous mode, information source collection module by Network card setup
It can monitor and capture and all by the network interface or flow through the data packet of the network interface card.
Due to traditional packet capture mechanism under high speed network there are performance bottleneck, " zero-copy " can be used
The capture work of thought realization data packet.The most basic thought of machine contracting system of catching based on zero-copy is exactly by data packet directly from net
Network device replication is to user's state space, and the system that reduces is called and the memory copying number of data packet, reduces the participation of CPU, makes
CPU is absorbed in the processing work of upper layer data packet.Concrete implementation is that area maps technology and direct memory are visited based on memory
Ask (DMA) technology.
Information collection module after reception of the data packet, can carry out layer-by-layer protocol analysis and be divided to the data packet captured
Analysis, the content analysis of each layer protocol of data packet is come out, thus the original document in restoring data packet.Principle is based primarily upon agreement
Analysis and reduction technique.
(1) protocal analysis, i.e., it is normative using the height of network protocol, pass through each layer to the data packet transmitted in network
Agreement is analyzed, and obtains the basic agreement information of data packet, parsing result is filled into corresponding data structure, for below
Stream recombination and using reduction prepare.If parsing data packet has occurred fragment or segmentation, also data packet is carried out
The recombination of fragment or segmentation.
(2) protocol assembly is to analyze on the basis of protocal analysis and stream recombination the agreement of application layer.It is not
Some data packet is analyzed, but the application layer conversation being made of volume of data packet is analyzed, that is, is being judged
On the basis of application layer protocol type, according to the exclusive workflow of various application layer protocols and feature (such as format, critical field
Deng), user's information of concern and application layer message text are extracted from session, after the processing such as unziping it, decode, general
Entire conversation procedure transmission is restored using data content.
3. Text Feature Extraction module
Text information extraction module, which is mainly responsible for, carries out in text the various files restored in information collection module
Hold and extracts.Semi-structured data and unstructured data two major classes can be classified as from the angle of the data type of file.Before
Person is one kind of structural data, however and do not meet the model structure of traditional relational, but include mark of correlation, be used to
Be layered and separated semantic primitive to field and record, the common file type for belonging to semi structured data have JSON,
HTML and XML etc..In addition a kind of unstructured data refers mainly to the data of not fixed structure, common various pictures, text
Shelves, audio/video etc. belong to unstructured data.
For the file type of semi-structured data, wherein JSON, HTML and XML etc. belong to standardized label language
Speech can generate corresponding resolver according to its specific language construction, to facilitate extraction text information.Wherein HTML determines
Justice has used the format of DTD (document type definition), which includes all elements being able to use and its hierarchical structure
With the definition of attribute.Parsing for HTML, mainly to the parsing of label, the result of parsing is one " analytic tree ", i.e., by
The tree that each attribute node and DOM element are constituted is one-to-one relationship between label and DOM.Resolving is broadly divided into
Two stages carry out,
(1) process of morphological analysis: i.e. input information is resolved to multiple labels by marking, and label mainly includes starting
Label, end label, attribute value and Property Name etc.;
(2) construct the process of dom tree: the output result in last stage is various HTML markups, these labels pass through label
Generator is transmitted to tree constructor, each that corresponding DOM element is marked to create after receiving respective markers, and is added to DOM
In tree.
And it is directed to the file type of most of unstructured data, although various documents, picture is without Uniform provisions
Type, and when specific to certain a kind of document therein, this kind of document can have general file structure.Such as PDF document etc.,
It is made of file header, end-of-file and file body.File header is mainly determined by information such as the version of document, size and dates;
End-of-file mainly saves the security information such as index and encryption;And file body is mainly made of various specific objects, including figure
Picture, text, various lines and font etc..Since the purpose of the module can be equally constructed to extract text information
It is used to carry out information extraction to file body for the resolver of particular document, by the definite position for getting each text in the page
It sets, to extract the information of corresponding text, lines, font size and title etc..And for the text envelope in picture
Breath extracts, can be by means of mature optical character identification (OCR) technology, by carrying out binaryzation, removal noise, school to image
Just and identification etc. processing carry out information extraction.
4. fingerprint generation module
Fingerprint generation module is mainly used to generate the finger print information of text file.The text refers to that Text Feature Extraction module mentions
The text file of taking-up.The method for generating finger print information generally refers to the Feature Words by extracting entire text, by these features
Set of words hashes to cryptographic Hash, and entire text is represented with these cryptographic Hash.
Traditional hash algorithm belongs to the scope of Pseudo-Random Number from principle, it is possible that two letters
The text of content difference very little is ceased, the finger print information of generation can have biggish difference, this can be to the text to be carried out later point
Analysis process causes tremendous influence;The finger print information of two texts obtained simultaneously by traditional hash algorithm, if fingerprint is not
It is identical to illustrate original content of text difference, and it is unable to get more information.
In order to solve problem above, the present invention has used for reference the thought of local sensitivity hash algorithm, and carries out on its basis
It improves.The feature of local sensitivity hash algorithm maximum is to keep the similitude between data, i.e., two poor in terms of content
After the text of other very little is by the part hash algorithm, difference locally only is being embodied, while as much as possible maintaining the two
Similitude.Its basic thought is the Feature Words for finding entire text, and the cryptographic Hash of these Feature Words is generated by hash algorithm,
Corresponding numeric string is generated according to the word frequency of Feature Words, the numeric string of all Feature Words is weighted, is added up and the behaviour such as dimensionality reduction
Obtain a numeric string.Its specific implementation process includes following components:
(1) it segments.One continuous sentence is split as to the process of individual word one by one according to specification.For English
For text, using space as delimiter between each word, therefore it need to only identify that participle process can be completed in space.However it is right
For most of Chinese text, text is mainly made of sentence, and sentence is most basic unit, without apparent delimiter.
It for the participle of Chinese, can be used based on character string Forward Maximum Method and the reversed maximum method for matching and being combined, by will be to
Since the character string of participle count matching word in both direction matched both direction with entry in machine dictionary respectively
Number takes maximum value in the two, and the phrase of the direction is as final word segmentation result.
(2) it denoises.After segmenting to text, the word segmentation result of either Chinese text or English text all can
In the presence of many noise words unrelated with text information correlation, as in English " a ", in " the " and Chinese " ",
" ", " good " etc., while there are also a large amount of punctuation marks.Entire text can be retained by removing these noise words and punctuation mark
Main information, while ensure that the accuracy of Feature Words.
(3) Feature Words are extracted.According to the above-mentioned word segmentation result after denoising, counts each phrase and go out in word segmentation result
Existing frequency, and all Feature Words are arranged according to descending, while result is exported with serial number+Feature Words+word frequency format.Due to
It is the extraction for single text, largely represents the content of entire text with the Feature Words that word frequency descending arranges.
(4) local sensitivity Hash.Above-mentioned Feature Words are become by hash algorithm by a string cryptographic Hash first, i.e., by counting
Word " 0 ", the numeric string of " 1 " composition;These cryptographic Hash are become into weighted number word string according to the word frequency of character pair word later, that is, are counted
Word is that the position of " 1 " adds word frequency, and number is that the position of " 0 " adds the opposite number of word frequency;Then all features after being weighted
The cryptographic Hash of word is added up to obtain a sequential value according to each;Finally this sequential value after cumulative is carried out at dimensionality reduction
It manages, i.e. each for the sequential value, label is on the position less than 0, and label is on the position greater than 0.This passes through dimensionality reduction
Sequential value after operation is just used as the finger print information of the text, is used for subsequent text analyzing.
5. text analysis model
Fingerprint database is the finger print information for storing urtext, herein urtext refer to need in advance it is to be protected
The text information of sensitive content.The fingerprint is generated again by above-mentioned Text Feature Extraction module with fingerprint generation module in advance.
It is arranged inside fingerprint database according to the filename of sensitive document, file size and file type, various ways is provided
It inquires related fingerprint, while file can be positioned by fingerprint.
The groundwork of text analysis model is by the finger of the sensitive information saved in the finger print information and fingerprint database
Line carries out the calculating of text similarity, use the result of text similarity as judge whether to have occurred sensitive information leakage according to
According to.If the text similarity between two finger print informations is higher than the secure threshold being previously set, i.e. the content letter of the two texts
It is largely same or similar for ceasing, then illustrates that the content information of the vital document saved is revealed.
Because finger print information is the feature vector generated after local sensitivity Hash by Feature Words, the finger print information is very
Entire text information can be represented in big degree, the similarity measurement between two text informations can be converted to by text envelope
Cease the similarity system design between the finger print information generated.Since finger print information is inherently feature vector after treatment, and
Similarity system design between vector can be embodied by the distance between vector.Distance between vector is weighed by Hamming distance
Amount counts the number of two isometric character string kinds of characters on corresponding position.The specific implementation of entire text analysis model
It is as follows:
(1) relevant fingerprint is inquired in fingerprint database according to the corresponding original text file size of finger print information;Or
Related fingerprint is inquired in fingerprint database according to the corresponding original text file type of finger print information;
(2) Hamming distance of each finger print information that the finger print information and inquiry that calculating generates obtain between the two;
(3) size for comparing each group Hamming distance and secure threshold finds that group distance less than or equal to secure threshold;
(4) confirm that sensitive information leakage, alert determine the sensitive document of leakage by fingerprint.
6. frame work process
For information collection module, specific calculating task is by packet capture, protocal analysis and protocol assembly group
At.First on needing the portal that monitors, catching machine contracting system and capture data packet using zero-copy, while will specifically catch
The fixed value that is dimensioned to of data packet is obtained, and this acquisition speed that can accelerate data packet is unlikely to missing data packet, it is in addition solid
Fixed data package size improves the efficiency of the reading and storage operation of data packet.Information collection module can be to the number of capture later
Protocal analysis is carried out according to packet and protocol assembly operates, with different after classifying for the data packet by different protocol encapsulations
Protocol analyzer parallel processing restores original document to corresponding data packet also according to different application layer protocols, thus divides
The degree of parallelism of protocal analysis and protocol assembly operation is not improved.
After information collection module restores file from the data packet of capture, multiple files for will being restored by output channel
It is transmitted to Text Feature Extraction module, while continuing to capture the data packet in network.Text Feature Extraction module is received in its input channel
After file, these files are divided by unstructured data and semi-structured data two major classes according to different file types, for
The file of semi-structured data type extracts text information by the processing such as morphological analysis and building dom tree, similarly for
The file of nonstructured data type is divided into image class and non-image two major classes, the former extracts text information using OCR technique, after
Person extracts text information using particular structured document parser, to realize that text is extracted in parallel processing in inside modules
Task.
The text information extracted can be transmitted to fingerprint generation module by output channel by Text Feature Extraction module.Fingerprint is raw
As long as receiving text information in its input channel at module, the work for carrying out fingerprint generation will be started, by segmenting, going
Final finger is obtained except noise word and punctuation mark, after the processes such as the Feature Words and the local sensitivity Hash that extract entire text
Line information.The finger print information of generation can be transmitted to text analysis model by output channel by fingerprint generation module later.
After text analysis model receives finger print information in its input channel, existed according to the original file information of this fingerprint
Similar set of fingerprint information therewith is searched in fingerprint database, carries out this fingerprint and finger print information collection in the inside modules later
The size between text similarity and secure threshold the two is compared in the calculating of text similarity between conjunction, to judge whether to occur
Sensitive information leakage.
The above modules are mutually indepedent, will not influence any intermediate state between each other at runtime, while will not
The problems such as in the presence of write-in conflict, intermodule carry out data-driven by input, output channel, and inside modules parallel processing, which calculates, appoints
Business.Because input, output channel are substantially the unidirectional fifo queues of non-obstruction, inside modules, as long as input channel
In there are data to meet the requirements, the case where which will work, be not in waiting, modules constitute on the whole
One assembly line that sensitive information leakage is measured in real time by data-driven.The sensitivity of data-driven proposed by the present invention
Modules in information leakage detection framework have used unified data stream programming model, while the leakage inspection of sensitive information
Survey is characterized eventually by the text similarity between fingerprint, is improved detection efficiency, is sufficiently addressed in high speed network environment
The processing speed of the data-bag lost, data packet that will appear in lower progress sensitive information leakage detection is slow and cannot judge in time
The problems such as response.
Detailed description of the invention
Fig. 1: frame construction drawing
Fig. 2: module programming illustraton of model
Specific embodiment
Exploitation environment of the invention is mainly based upon a PC host.Wherein, the CPU of PC machine is Intel (R) Core (TM)
I5-4570CPU 3.20GHz, GPU are Intel (R) Haswell Desktop, inside save as 4GB RAM, hard disk size is
204.2G, system are ubuntu 16.04LTS, and 64 bit manipulation systems, network interface card rate is 1000Mb/s.
The calculating task of modules mainly uses python, the language developments such as bash in the present invention.The capture of data packet
And file extracts the FileAnalysis Framework by means of bro tool.Python version is 2.7.12, bash version
For 4.3.48.
Experimental data is the data packet that test machine captures in file download, transmission process and is used on internet
The network packet of secure context practice and research.
Primary operational is broadly divided into three parts, and first part is that text information extracts part, and second part is finger print information
Generating portion, Part III are text analyzing part.
1. text information extracts part
(1) algorithm description
Algorithm input: Files
Algorithm output: T
Illustrate: Files is the file set that information collection module is restored from data packet, and T is to extract from file
Text information set.
Algorithm steps:
1) semi-structured data is divided into the file in file set and two class of unstructured data is handled;
2) for the file of semi-structured data type, input file is parsed by multiple marks by the process of morphological analysis
Then these marker configurations are counted up to the reading of pairs of mark information finally by traversal whole, in turn at a dom tree by note
Extract text information;
3) for the file of unstructured data types, it is divided into image and non-image two types is handled;
4) it for the file of image type, is handled using optical character recognition technology (OCR), mainly passes through image two-value
Change, remove the processing such as noise, correction and identification to obtain text information;
5) for remaining unstructured data types file, text is extracted using the resolver of particular file types.
2. finger print information generating portion
(1) algorithm description
Algorithm input: T, N, W, len
Algorithm output: F
Illustrate: T is the set of the text information extracted from file, and N is the noise word and punctuate symbol of manual creation
Number set, W be Chinese dictionary constitute dictionary, len be dictionary in phrase maximum length, F be generation finger print information
Set.
Algorithm steps:
1) word segmentation processing is carried out one by one to the text in text information set, the text after generating participle;
2) according to the noise set of words of offer, noise reduction process is carried out to the text after participle, removes the noise word in text
And punctuation mark;
3) number that each phrase occurs is counted in the text after noise reduction, and phrase is subjected to descending according to frequency of occurrence
Arrangement, obtains the set of Feature Words;
4) each Feature Words are become by cryptographic Hash by hash algorithm;
5) cryptographic Hash be weighted, added up, generating final finger print information after the processing such as dimensionality reduction.
Its pseudocode is as follows:
3. text analyzing part
(1) algorithm description
Algorithm input: Finger_1, fingerDatabase, Thresholds
Algorithm output: Result
Illustrate: Finger_1 is the finger print information to be detected generated, and fingerDatabse is finger print information storehouse,
Thresholds is the secure threshold that text similarity compares setting, and Result is text analyzing as a result, being used to indicate whether
Sensitive information leakage has occurred.
Algorithm steps:
1) it is searched in fingerprint database and set of fingerprint information similar in fingerprint to be detected;
2) Hamming distance of each finger print information that the finger print information and inquiry that calculating generates obtain between the two;
3) size for comparing each group Hamming distance and secure threshold finds that group distance less than or equal to secure threshold;
4) confirm sensitive information leakage, alert.
Its pseudocode is as follows:
Claims (6)
1. a kind of sensitive information leakage detection framework of data-driven, characterized by comprising:
(1) information collection module, for being captured to the data packet transmitted on the portal, to the data packet after capture into
Row protocal analysis and parsing are to restore original data content;
(2) Text Feature Extraction module, for carrying out Text Feature Extraction to the data content restored in information collection module, according to not
Same data type extracts text respectively;
(3) fingerprint generation module, for the text extracted to be generated corresponding finger print information by local sensitivity Hash;
(4) fingerprint database, for storing the finger print information for needing sensitive content to be protected in advance;
(5) text analysis model, the fingerprint of the sensitive information for being saved in the finger print information and fingerprint database by generation into
The calculating of row text similarity uses the result of text similarity as the foundation for judging whether to have occurred sensitive information leakage.
2. the sensitive information leakage detection framework according to claim 1 by data-driven, it is characterised in that should detect
The information collection module of frame can carry out information to the network of monitoring and collect and therefrom restore data:
(1) machine contracting system is caught using zero to the data packet to circulate on the network node of monitoring to capture;
(2) data packet of capture is subjected to protocal analysis according to its transport protocol, obtains different types of data packet;
(3) by stream recombination and protocol assembly, original data content is restored from data packet.
3. the sensitive information leakage detection framework according to claim 1 by data-driven, it is characterised in that this detection block
The Text Feature Extraction module of frame can be divided into semi-structured and unstructured two types to the data restored and carry out Text Feature Extraction:
(1) it for the partly-structured data restored, is parsed by morphological analysis and building two stages of dom tree entire
Data recursively extract text from dom tree;
(2) for the data of the structuring restored, it is divided into image class and two class of non-image class is respectively processed, the former uses
Optical character identification extracts text, and the latter extracts text by means of the resolver of particular file types.
4. the sensitive information leakage detection framework according to claim 1 by data-driven, it is characterised in that this detection block
The fingerprint generation module of frame can generate the finger for representing entire text using local sensitivity hash algorithm to the text extracted
Line, implementation step are as follows:
(1) entire content of text is split into individual phrase one by one by character string maximum matching algorithm;
(2) the various meaningless noise words and punctuation mark in text are removed;
(3) descending arrangement is carried out according to the frequency that phrase occurs to the text after denoising, extracts the feature for representing entire text
Word;
(4) by carrying out Hash operation to Feature Words, the cryptographic Hash of each Feature Words is weighted, is added up and dimensionality reduction etc.
Final finger print information is obtained after reason.
5. the sensitive information leakage detection framework according to claim 1 or 4 by data-driven, it is characterised in that this detection
The fingerprint database of frame stores the finger print information for needing sensitive content to be protected in advance, and fingerprint equally generates mould by fingerprint
Block generates, and is arranged inside the fingerprint database according to the filename of sensitive content, file size and file type, can be with
It inquires related fingerprint in several ways, while file can be positioned by fingerprint.
6. according to claim 1 or 5 by the sensitive information leakage detection framework of data-driven, it is characterised in that this detection
The text analysis model of frame quickly can judge whether that sensitive information leakage has occurred by the text similarity between fingerprint:
(1) finger print information to be detected generated according to fingerprint generation module searches the fingerprint of same type in fingerprint database,
Obtain close set of fingerprint information;
(2) Hamming distance between fingerprint to be detected and same type fingerprint set the two is calculated;
(3) compare the size between each group Hamming distance and secure threshold distance, wherein one group of Hamming distance is less than or equal to if it exists
Secure threshold distance, the content for representing this group of finger print information is essentially identical, illustrates in original text corresponding to fingerprint to be detected
Appearance is exactly shielded sensitive information, also just detected the generation of sensitive information leakage.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910030202.3A CN109766525A (en) | 2019-01-14 | 2019-01-14 | A kind of sensitive information leakage detection framework of data-driven |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910030202.3A CN109766525A (en) | 2019-01-14 | 2019-01-14 | A kind of sensitive information leakage detection framework of data-driven |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109766525A true CN109766525A (en) | 2019-05-17 |
Family
ID=66453859
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910030202.3A Pending CN109766525A (en) | 2019-01-14 | 2019-01-14 | A kind of sensitive information leakage detection framework of data-driven |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109766525A (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110377977A (en) * | 2019-06-28 | 2019-10-25 | 南方电网科学研究院有限责任公司 | Sensitive information leakage detection method and device and storage medium |
CN110618923A (en) * | 2019-09-18 | 2019-12-27 | 阿里巴巴集团控股有限公司 | Data monitoring method and system |
CN111159703A (en) * | 2019-12-31 | 2020-05-15 | 奇安信科技集团股份有限公司 | Virtual machine data leakage detection method and device |
CN111177362A (en) * | 2019-12-30 | 2020-05-19 | 深圳市梦网科技发展有限公司 | Information processing method, device, server and medium |
CN111209373A (en) * | 2020-01-07 | 2020-05-29 | 北京启明星辰信息安全技术有限公司 | Sensitive text recognition method and device based on natural semantics |
CN111629027A (en) * | 2020-04-10 | 2020-09-04 | 云南电网有限责任公司信息中心 | Trusted file storage processing method based on block chain |
CN111917718A (en) * | 2020-06-24 | 2020-11-10 | 武汉绿色网络信息服务有限责任公司 | Personal information leakage monitoring method and device |
CN112733188A (en) * | 2021-01-13 | 2021-04-30 | 航天晨光股份有限公司 | Sensitive file management method |
CN114338155A (en) * | 2021-12-28 | 2022-04-12 | 四川邦辰信息科技有限公司 | Network privacy protection method and system based on multi-dimensional fingerprint confusion |
CN115499156A (en) * | 2022-07-29 | 2022-12-20 | 天翼云科技有限公司 | Website background information leakage detection method, electronic device and storage medium |
CN116795793A (en) * | 2023-06-26 | 2023-09-22 | 珠海精实测控技术股份有限公司 | Data interaction method and device based on standardized file and storage medium |
CN118227684A (en) * | 2024-05-23 | 2024-06-21 | 中国电子科技集团公司第三十研究所 | Data behavior tracing method based on multidimensional discrete data fingerprint |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101795230A (en) * | 2010-02-23 | 2010-08-04 | 西安交通大学 | Network flow recovery method |
CN104363240A (en) * | 2014-11-26 | 2015-02-18 | 国家电网公司 | Unknown threat comprehensive detection method based on information flow behavior validity detection |
CN105956482A (en) * | 2016-04-15 | 2016-09-21 | 宝利九章(北京)数据技术有限公司 | Method and system for data leakage protection |
CN107577939A (en) * | 2017-09-12 | 2018-01-12 | 中国石油集团川庆钻探工程有限公司 | Data leakage prevention method based on keyword technology |
-
2019
- 2019-01-14 CN CN201910030202.3A patent/CN109766525A/en active Pending
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101795230A (en) * | 2010-02-23 | 2010-08-04 | 西安交通大学 | Network flow recovery method |
CN104363240A (en) * | 2014-11-26 | 2015-02-18 | 国家电网公司 | Unknown threat comprehensive detection method based on information flow behavior validity detection |
CN105956482A (en) * | 2016-04-15 | 2016-09-21 | 宝利九章(北京)数据技术有限公司 | Method and system for data leakage protection |
CN107577939A (en) * | 2017-09-12 | 2018-01-12 | 中国石油集团川庆钻探工程有限公司 | Data leakage prevention method based on keyword technology |
Cited By (18)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110377977A (en) * | 2019-06-28 | 2019-10-25 | 南方电网科学研究院有限责任公司 | Sensitive information leakage detection method and device and storage medium |
CN110618923B (en) * | 2019-09-18 | 2023-09-15 | 创新先进技术有限公司 | Data monitoring method and system |
CN110618923A (en) * | 2019-09-18 | 2019-12-27 | 阿里巴巴集团控股有限公司 | Data monitoring method and system |
CN111177362A (en) * | 2019-12-30 | 2020-05-19 | 深圳市梦网科技发展有限公司 | Information processing method, device, server and medium |
CN111159703A (en) * | 2019-12-31 | 2020-05-15 | 奇安信科技集团股份有限公司 | Virtual machine data leakage detection method and device |
CN111209373A (en) * | 2020-01-07 | 2020-05-29 | 北京启明星辰信息安全技术有限公司 | Sensitive text recognition method and device based on natural semantics |
CN111629027A (en) * | 2020-04-10 | 2020-09-04 | 云南电网有限责任公司信息中心 | Trusted file storage processing method based on block chain |
CN111917718A (en) * | 2020-06-24 | 2020-11-10 | 武汉绿色网络信息服务有限责任公司 | Personal information leakage monitoring method and device |
CN111917718B (en) * | 2020-06-24 | 2023-04-07 | 武汉绿色网络信息服务有限责任公司 | Personal information leakage monitoring method and device |
CN112733188A (en) * | 2021-01-13 | 2021-04-30 | 航天晨光股份有限公司 | Sensitive file management method |
CN112733188B (en) * | 2021-01-13 | 2023-09-22 | 航天晨光股份有限公司 | Sensitive file management method |
CN114338155A (en) * | 2021-12-28 | 2022-04-12 | 四川邦辰信息科技有限公司 | Network privacy protection method and system based on multi-dimensional fingerprint confusion |
CN114338155B (en) * | 2021-12-28 | 2024-04-30 | 四川邦辰信息科技有限公司 | Network privacy protection method and system based on multidimensional fingerprint confusion |
CN115499156A (en) * | 2022-07-29 | 2022-12-20 | 天翼云科技有限公司 | Website background information leakage detection method, electronic device and storage medium |
CN115499156B (en) * | 2022-07-29 | 2024-06-07 | 天翼云科技有限公司 | Website background information leakage detection method, electronic equipment and storage medium |
CN116795793A (en) * | 2023-06-26 | 2023-09-22 | 珠海精实测控技术股份有限公司 | Data interaction method and device based on standardized file and storage medium |
CN116795793B (en) * | 2023-06-26 | 2024-06-11 | 珠海精实测控技术股份有限公司 | Data interaction method and device based on standardized file and storage medium |
CN118227684A (en) * | 2024-05-23 | 2024-06-21 | 中国电子科技集团公司第三十研究所 | Data behavior tracing method based on multidimensional discrete data fingerprint |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109766525A (en) | A kind of sensitive information leakage detection framework of data-driven | |
CN110516067B (en) | Public opinion monitoring method, system and storage medium based on topic detection | |
CN108737423B (en) | Phishing website discovery method and system based on webpage key content similarity analysis | |
CN106202561B (en) | Digitlization contingency management case base construction method and device based on text big data | |
WO2022041406A1 (en) | Ocr and transfer learning-based app violation monitoring method | |
CN112765603B (en) | Abnormity tracing method combining system log and origin graph | |
CN111585955B (en) | HTTP request abnormity detection method and system | |
CN110413787B (en) | Text clustering method, device, terminal and storage medium | |
CN108875366A (en) | A kind of SQL injection behavioral value system towards PHP program | |
CN109194677A (en) | A kind of SQL injection attack detection, device and equipment | |
Suo et al. | A simple and robust correlation filtering method for text-based person search | |
CN111814192B (en) | Training sample generation method and device and sensitive information detection method and device | |
US20050050086A1 (en) | Apparatus and method for multimedia object retrieval | |
CN102117339A (en) | Filter supervision method specific to unsecure web page texts | |
WO2022127259A1 (en) | Data cleaning method, apparatus and device, and storage medium | |
CN113986864A (en) | Log data processing method and device, electronic equipment and storage medium | |
CN111460803B (en) | Equipment identification method based on Web management page of industrial Internet of things equipment | |
Liu et al. | Scene text detection with fully convolutional neural networks | |
WO2020101479A1 (en) | System and method to detect and generate relevant content from uniform resource locator (url) | |
CN112328792A (en) | Optimization method for recognizing credit events based on DBSCAN clustering algorithm | |
CN114443600A (en) | Data log analysis system and method | |
CN111581057B (en) | General log analysis method, terminal device and storage medium | |
Setianto et al. | Gpt-2c: A gpt-2 parser for cowrie honeypot logs | |
CN111383660A (en) | Website bad information monitoring system and monitoring method thereof | |
CN113343051B (en) | Abnormal SQL detection model construction method and detection method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190517 |